The problem in vqa_dataset.py

Karine-Huang / T2I-CompBench

[Neurips 2023] T2I-CompBench: A Comprehensive Benchmark for Open-world Compositional Text-to-image Generation

https://arxiv.org/pdf/2307.06350.pdf

MIT License

168 stars 5 forks source link

The problem in vqa_dataset.py #15

Open k101w opened 4 months ago

k101w commented 4 months ago

In the vqa_dataset.py, I don't understand when the split != 'train', self.annotation will be set to 'color_test.json'. Should we modify this name manually when we are testing 'shape' or 'texture'? WechatIMG240

Karine-Huang commented 4 months ago

Hello! You don't need to modify the name manually; it doesn't impact the results. We've still made the change from 'color_test.json' to 'vqa_test.json'.